Merge #4674

4674: 1586881: Fix high error rate on glean.baseline.duration r=mdboom a=mdboom

There is a race condition in how the TimingManager is used here.
When stopping a timer, we dispatch a task to actually record the value in the
timespan, and only reset the TimingManager in that async task after the
recording, which requires I/O, is complete.

If the timespan is started in the meantime, an error will be recorded, rather
than safely starting a new timespan measurement.

This change could be considered on the principle (followed elsewhere), that
we do timing management and measurement synchronously, and only recording
asynchronously.
Co-authored-by: Michael Droettboom <mdboom@gmail.com>

components/service/glean/src/main/java/mozilla/components/service/glean/private/TimespanMetricType.kt

@@ -83,11 +83,11 @@ data class TimespanMetricType(
                 @Suppress("EXPERIMENTAL_API_USAGE")
                 Dispatchers.API.launch {
                     TimespansStorageEngine.set(this@TimespanMetricType, timeUnit, elapsedNanos)
-
-                    // Reset the timerId.
-                    timerId = null
                 }
             }
+
+            // Reset the timerId.
+            timerId = null
         }
     }